Latest / Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™ / Agentic AI for Back-Office Workflows — ADA AI

Transcript

Agentic AI for Back-Office Workflows — ADA AI · Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™

• 0:00
  If you're a founder, operator or executive, here's the
• 0:03
  uncomfortable truth. Your back office is stuck in
• 0:07
  2015 if you're drowning in repetitive
• 0:11
  work, operational inefficiencies and processes that
• 0:15
  break every time you try to scale Today's guest
• 0:18
  believes the future of work will not be AI
• 0:22
  assistants, but AI employees,
• 0:25
  autonomous agents that think, plan and
• 0:28
  execute real workflows. Dr. Oliver
• 0:31
  Lugosch built Razor Group to more than 700 million in
• 0:35
  annual revenue, managing global operations at
• 0:39
  massive scale. He's now building ADA
• 0:42
  AI, or ADA AI as you can say in German, where
• 0:46
  AI employees are already replacing entire back office
• 0:50
  workflows. In this episode we we break
• 0:53
  down what AI employees can actually do today,
• 0:58
  where they fail, and how founders can employ their first
• 1:01
  agents in under 30 days. Get ready.
• 1:05
  This conversation will fundamentally change how you think about
• 1:08
  operations.
• 1:15
  Welcome to Startuprad IO,
• 1:19
  your podcast and YouTube blog covering the German
• 1:23
  startup scene with news, interviews and
• 1:26
  live events.
• 1:31
  Today we're joined by Dr. Oliver Glugosz, a founder
• 1:34
  who has lived through one of the most operationally intense
• 1:38
  journeys in Germany's startup ecosystem. Oliver co
• 1:41
  founded the Razer Group, an e commerce powerhouse that
• 1:45
  scaled to over 700 million euros in annual revenue,
• 1:49
  navigated global supply chains and estimated executed M and A
• 1:53
  at the pace that would break most companies. That
• 1:56
  experience gave him a firsthand look into the limits of human
• 2:00
  only operations. The arrows, the repetitions, the
• 2:04
  scalability ceiling. And it led him to a bold
• 2:07
  idea. What if startups didn't hire more humans
• 2:11
  for repetitive work, but deploy
• 2:15
  AI employees instead? Today all Oliver is a co
• 2:18
  founder of ADA AI, one of
• 2:22
  Europe's emerging leaders in agentic AI, a
• 2:26
  new category where autonomous agents don't just
• 2:29
  assist, but act, plan and operate
• 2:33
  independently. These AI employees handle full fledged
• 2:37
  workflows from data processing to back office tasks,
• 2:40
  customer operations, and complex decision paths that
• 2:44
  previously required entire teams. In this
• 2:47
  interview we'll explore the real limits of traditional automation,
• 2:51
  how AI employees differ from chatbots and LLM
• 2:55
  assistants, where autonomous agents shine and
• 2:58
  where they break, what founders must know before deploying their first
• 3:02
  AI employee and how AgentIC AI will reshape the
• 3:06
  startup workforce over the next five years. They
• 3:10
  that was a long introduction. Oliver, welcome. Thank you very
• 3:13
  much. Thanks for having me. It is totally my pleasure.
• 3:19
  I'm very, very curious about those AI employees,
• 3:23
  especially having a company where I still do
• 3:28
  maybe more work than needed manually. It is always something
• 3:31
  I had an eye on but could never really get the time
• 3:35
  to do it. That is something of real interest and I
• 3:39
  do believe A lot of founders here will, will really,
• 3:42
  really listen carefully. If they can do something like their
• 3:46
  monthly tax filings, that will be just amazing. But
• 3:49
  first, let us start. You scaled Razor Group into one of
• 3:53
  Germany's largest E commerce operations. What was the moment where
• 3:57
  you realized the traditional human only back office model simply
• 4:01
  wouldn't scale anymore? Wow. Very
• 4:04
  good question. Very good question. Let me think back to the time when we founded
• 4:08
  Razor Group. I did this with three other founders
• 4:12
  back in summer of 2020. It was
• 4:16
  just a few months after the pandemic hit. So everything
• 4:20
  was changing, everything was in a turmoil. Things
• 4:24
  were changing at a rapid pace.
• 4:27
  And so we started our journey
• 4:31
  based out of Berlin, getting in touch with e
• 4:35
  commerce companies, small e commerce companies that we first
• 4:39
  had to explain to, you know, that you can actually sell
• 4:42
  your business, right? You can find a buyer and sell your business
• 4:47
  to somebody else and have a, you know, life changing
• 4:50
  exit event. That was what we started with
• 4:54
  and that was, was more, that is what most of Razer was
• 4:58
  focused on, right? Getting these deals, finding these
• 5:01
  companies that you could acquire. My role as
• 5:05
  CEO and co founder was to think about the operations before we
• 5:09
  had any operations, right? Everybody was chasing these deals and these
• 5:12
  companies. And I was thinking, okay, once we close those deals, once we buy
• 5:16
  these companies, what do we do? How do we operate, right?
• 5:20
  Starting obviously with a small team, having a few people on board,
• 5:25
  the first deals started to come in. I think we acquired our first
• 5:29
  company in October of 2020, like two, three months after founding
• 5:32
  the, the company. And it was all,
• 5:36
  you know, setting up the absolute basics, setting up structures,
• 5:40
  setting up responsibilities, setting up
• 5:43
  processes. And I very clearly remember this first company
• 5:47
  that we acquired. Everything was hands on, right? Everything
• 5:50
  was Excel files, very, very manual. Everything
• 5:54
  was double checked by a, by a colleague, human, human
• 5:58
  interventions. Nothing was, was really system
• 6:01
  based. We did know that we would grow very rapidly.
• 6:05
  So at that time we already checked what
• 6:09
  ERP system do we want to use, right? We already thought about scalability,
• 6:13
  had that process running and ultimately even went live with
• 6:16
  our ERP at the turn of the, of the year, just
• 6:20
  like five, six months after founding the company, we were live with our
• 6:24
  erp. We, we chose Oracle netsuite back then. And in
• 6:28
  that fall of after the first acquisition,
• 6:31
  the second came a third came right, and there was still a pace
• 6:35
  that you could manage, right? You hire another colleague, another
• 6:39
  specialist for logistics, another one maybe for E commerce
• 6:42
  operations. But when this wave of
• 6:46
  acquisitions came around, where we heard,
• 6:49
  okay, in January, we might do four to six
• 6:53
  acquisitions that was my oh shit
• 6:57
  moment, pardon my French. That was
• 7:00
  exactly. That was the moment where I thought, okay, you can never
• 7:04
  hire at high quality at that pace, right? How do you do
• 7:08
  this? And so it was the beginning of
• 7:12
  thinking even more in terms of scalability, even more
• 7:15
  in terms of automation. Standardization. The first step for us
• 7:19
  was to standardize the integration process, right?
• 7:23
  Understand what are the basic steps that you need to do, in
• 7:27
  what order, how do you standardize the entire process so that you can
• 7:30
  handle four to six integrations per month.
• 7:35
  And that was really the first piece that we had to get in line
• 7:39
  to. Then also in second step, think about the operations, right? Once
• 7:43
  you integrate that company, it is part of your landscape, it's part of your
• 7:47
  IT infrastructure, of your system, process infrastructure, team
• 7:50
  infrastructure, then how do you operate it? And so we
• 7:54
  always thought in these waves, starting with the integration all the way to
• 7:58
  operations, and then improvement, continuous improvement,
• 8:01
  implementing more and more standardization, operations,
• 8:05
  automation and all of these things. I was
• 8:09
  thinking about a company that acquires a lot of other company,
• 8:12
  aggregates them. I would instantly have in mind
• 8:16
  like one big machine. And you take. Get
• 8:20
  rid of as many individual processes as possible and
• 8:24
  add all those processes to as
• 8:28
  efficient working central machine as possible.
• 8:32
  Absolutely. Try to standardize as much as possible, right? Because if
• 8:35
  you standardize, you can repeat. It's a blueprint, right? This is the
• 8:39
  plan, go execute. You can use some basic
• 8:43
  automation. We are not talking about LLMs yet, right? LLMs have not been
• 8:46
  anywhere in mainstream at that time, 2020. But it was
• 8:50
  more of if then connections, right? If there is an email coming
• 8:54
  in, then execute that. Or if somebody pulls a project from a certain
• 8:57
  stage to the next one, you automatically send that set of
• 9:01
  documents to a certain person, right? So yeah. And this
• 9:04
  IFTT logic was already around in the 90s in
• 9:08
  Excel. Exactly. Why do you think the back office and
• 9:11
  operational processes remain so repetitive,
• 9:16
  manually error prone and resistant to automation in.
• 9:20
  Even in tech companies, there's a certain degree
• 9:24
  to which they need to be in certain companies. But
• 9:28
  why even in tech companies, you do everything in front
• 9:32
  of us to apply LLMs, to apply agents, to make it
• 9:36
  as easy as possible. And then in the back office
• 9:40
  you start working paper basement stamps. Yeah,
• 9:44
  very good question. I think it has two main components,
• 9:47
  really. One component is still of
• 9:51
  technical nature, right? Before LLMs,
• 9:55
  you could not automate the process of understanding communication,
• 9:58
  right? How do you automate an email communication process? Right.
• 10:02
  There was no tool that could understand, for example, a
• 10:05
  supplier's message. And then reason think
• 10:09
  through what the appropriate answer is, and then put that answer into an
• 10:13
  email and send it. There was just no technology. Right. And that technology, these
• 10:17
  large language models, have now only been around really for a couple of
• 10:20
  quarters. Right. So it's still a novelty in terms of technology, but
• 10:24
  it is available. It is feasible nowadays, so
• 10:27
  that problem is solved. I think the other side of the coin is the
• 10:31
  topic of trust. Trust and accountability. Right.
• 10:35
  Because when it comes to decisions and to actions, even if
• 10:39
  it's just as small as a communication to a customer, to
• 10:43
  a supplier, companies tend to still trust
• 10:46
  humans more than machines. Right. And if there is a decision
• 10:50
  that needs to be taken, for example, on changing delivery dates, on
• 10:54
  accepting a certain change, then structures and
• 10:57
  organizations rather trust a human to take that call because that human
• 11:01
  would be accountable for the outcome. And how do you hold a machine accountable to
• 11:05
  something? Right. I think that's the second element that. I
• 11:08
  think that, that, that's an important question everybody needs to
• 11:12
  answer in the future. It's, it's, I do believe the same question. How do you
• 11:16
  hold a company accountable? How can you hold
• 11:19
  an AI agent accountable? It will more or
• 11:23
  less end up to. To be the accountability
• 11:27
  of some human either in charge of those agents or the person
• 11:31
  who coded those agents. What do you think?
• 11:35
  I agree. I think it's a process that we'll go through
• 11:39
  finding exactly how these things are settled. But we do have examples
• 11:42
  nowadays. Right. So who is responsible ever? I'd say
• 11:46
  classical automation fails. Right? It's the
• 11:50
  party who set up that automation or who
• 11:53
  agreed to maintain it, to oversee it.
• 11:57
  Right. You clearly define this upfront. And I think we're moving into a
• 12:01
  world where you need to define it between the parties.
• 12:04
  And it's typically something that needs to develop over
• 12:08
  time. Kind of what is the standard that everybody agrees to
• 12:12
  and how is that all figured out? I think that's a process that
• 12:16
  we are in the middle of seeing the development
• 12:20
  of. I'm very sure a lot of people who
• 12:24
  are deploying or thinking about deploying
• 12:27
  AI agents are now thinking, huh, that's a good
• 12:31
  question. Never thought of that. What was
• 12:34
  like the spark moment that led you to the idea of
• 12:38
  an AI employee autonomous
• 12:42
  agent that can run a full workflow?
• 12:46
  It really was a thought
• 12:50
  that came up during the entire journey of Razor Group. Right.
• 12:53
  That journey for me was roughly five years long. I just explained
• 12:57
  what happened in the first few months. And over the next
• 13:01
  quarters, we built that organization. It grew and grew and grew.
• 13:05
  And we always saw that there is a lot of manual work
• 13:09
  required to really keep things running. Right? You have these,
• 13:13
  I would say that API of anything which is human
• 13:17
  led. Humans are the API of anything because they understand things,
• 13:20
  they can verify things, you can teach them things.
• 13:24
  And so they are bridging what classical
• 13:28
  systems like ERPs cannot solve, right? For
• 13:31
  example, the interaction. And over these five years, as the,
• 13:35
  as the organization grew, every, every now and then I was thinking
• 13:38
  about, wow, this, this, there must be a different way, there must be a better
• 13:42
  way, there must be a way to automate also these processes that are
• 13:46
  highly repetitive that you can put into an
• 13:49
  sop, a standard operating procedure, but the technology
• 13:53
  just wasn't there. And then in 2024 when,
• 13:57
  you know, 01 and 4O from ChatGPT, from OpenAI
• 14:01
  came out, that was really a pivotal moment for
• 14:04
  me and I think also for a lot of other people,
• 14:08
  seeing the potential that comes
• 14:12
  with these models of understanding, communication, of understanding
• 14:15
  context, of understanding, documents that had this
• 14:19
  aha moment, wow, now we have the technology that is
• 14:23
  reliable enough to do these things, to do these
• 14:26
  previously exclusively human processes now
• 14:29
  in a machine. And that was the point where we said, let's go ahead,
• 14:33
  let's see what we can do with that technology. And that was
• 14:37
  really the spark for us to embark on the journey.
• 14:41
  And that actually leads us to the next question. What did you do
• 14:45
  with those AI employees from Eraser Group?
• 14:48
  Experience which operational bottleneck
• 14:52
  with the biggest cost sync or scalability
• 14:55
  barrier? And how would an AI
• 14:59
  employee have solved it? Razer is an
• 15:03
  E commerce business and Razer sells
• 15:07
  everyday goods, right? And so the
• 15:10
  procurement of these goods is very
• 15:14
  fragmented. We had a lot of different suppliers from
• 15:18
  all around the globe, most of which in fact came from, from
• 15:21
  Asia. And the coordination with all
• 15:25
  of these suppliers always was a very, very human
• 15:28
  driven task, a very manual task. Because a lot of suppliers weren't that
• 15:32
  large, right? They didn't have massive, massive factories. They
• 15:36
  may not have had, you know, large teams and
• 15:40
  also IT teams that work on integrations,
• 15:43
  standardized interfaces, etc. But they, they had no
• 15:47
  API. You just could not put your order in your ERP and
• 15:50
  went over, you had to put it somewhere in your erp, then double
• 15:54
  it up, send an email, getting confirmation and so on and so forth.
• 15:58
  A lot of back and forth, right? Absolutely, 100%,
• 16:01
  a very, very manual process. And that was
• 16:05
  the field where we figured, wow, we have a lot of volume, right?
• 16:09
  Every single product that we buy and then sell to
• 16:13
  customers goes through that process. We had a lot of fragmentation,
• 16:17
  right? I think about 500 suppliers or even more. A lot of
• 16:20
  different languages at play, right? Sourcing from China, from Vietnam,
• 16:25
  from India, from all around the globe and all of that variety.
• 16:29
  And we figured, well, this is something that we want to take up first
• 16:33
  and set up a trial. And that trial, taking
• 16:36
  care of that supplier communication and coordination, right? The
• 16:40
  exchange of, update of dates, update of quantities,
• 16:44
  etc. Tackling that that was our first use case that we
• 16:47
  approached and we just saw the impact that you can
• 16:51
  create and that was, that was a real eye opener for us.
• 16:56
  Before we get into the question about your assumption about
• 17:00
  starting ADA AI, how did you get the name?
• 17:05
  Very good question. Sreshta, who is
• 17:09
  a co founder of mine, she was also
• 17:13
  a co founder at Razor Group CTO and is now also co
• 17:16
  founder at ADA AI. She's, she
• 17:20
  studied computer science in Stanford. So she's a techie through and through.
• 17:25
  And I remember it was her idea that we name
• 17:29
  ADA ADA because of ADA Loveless, which I
• 17:33
  learned was the first computer programmer.
• 17:36
  And you know, sounds good,
• 17:40
  it has a good ring to it. And so I think we went on board
• 17:43
  with it and like the name so far. When you then started
• 17:47
  ada, what was your assumption about
• 17:50
  agentic AI? And what assumption turned out to be
• 17:54
  completely wrong? If you go through,
• 17:58
  you know, LinkedIn, the media, any business related
• 18:03
  social media out there, you always get the
• 18:06
  statement, do vertical AI, do highly
• 18:09
  specialized AI just become the best AI for
• 18:13
  customs declarations, become the best AI for
• 18:18
  indirect procurement? And that was something that
• 18:21
  we heard, that we listened to, but that ultimately
• 18:25
  for us turned out to not be the right
• 18:29
  approach. Because when we started to speak to
• 18:32
  customers, basically asking them, hey, where do you have
• 18:36
  repetitive manual work? Where do you have processes
• 18:39
  that have not been automated yet? Mostly because they include a lot of
• 18:43
  communication, a lot of language, a lot of
• 18:47
  unstructuredness in the data. The responses that we got
• 18:51
  were all over the place, all over the entire supply chain from
• 18:55
  procurement to distribution,
• 18:58
  logistics, invoice processing,
• 19:02
  sales from all parts of the business. And we started to look
• 19:06
  into these use cases and started to look, okay, what does it take
• 19:10
  to now build an automation that works end to end? And that
• 19:14
  really takes that burden of manual work off
• 19:17
  of the team's shoulders. And it turns out that all of
• 19:21
  these are similar in terms of what
• 19:25
  you need to make them work in the front end. When you look at
• 19:29
  it at face value, these use cases are very, very different, right?
• 19:33
  One use case may be supplier communication. A supplier
• 19:36
  notifies you of delays, or a supplier Notifies you of
• 19:40
  change of quantity, right? It's completely different when you compare it
• 19:44
  to, I'm processing order in a, in a food
• 19:48
  business, right? I'm processing small individual orders from
• 19:51
  small retailers, right? I get emails, I need to interpret the email, need to
• 19:55
  pull out the items from the email and put them into the erp. Two
• 19:59
  very different use cases. But if you look under the hood,
• 20:03
  what you need to make them work is really comparable because
• 20:06
  what are the elements that you need? You need to be able to understand
• 20:10
  communication, number one. Number two, you need to
• 20:14
  match certain data in an unstructured way to another
• 20:17
  set of data, right? Because people don't use exact
• 20:21
  namings of products, people don't use exact phrases, but
• 20:25
  you have to do what we call fuzzy match. You need to be integrated
• 20:29
  into these different systems. You need to have an
• 20:32
  engine that understands context. You need to
• 20:36
  have a user interface, right? Because you need to think, what does the team do?
• 20:39
  The team needs to interact with the AI. And there are a couple more elements
• 20:43
  and modules you could call them. And we identified that
• 20:47
  these by themselves are quite repeatable.
• 20:51
  They are quite, you know, you just need to arrange them a little
• 20:55
  bit, you need to readjust them a little bit, but you can reuse them. And
• 20:59
  that was something that came out through the interaction with other
• 21:02
  customers, listening to them, going through the process of
• 21:06
  figuring out how do you build these automations? That then told us, well,
• 21:09
  actually just being vertical, being extremely narrow
• 21:13
  is not what yields the best result for the customer.
• 21:17
  But being able to think horizontally across the entire
• 21:20
  supply chain, across the entire, entire process landscape that you
• 21:24
  have in a company is really what, what turns
• 21:28
  out to create the most value for the customer.
• 21:33
  A lot of companies out there really love the idea of
• 21:37
  automation, but feel the real autonomy.
• 21:41
  Where do you see the biggest misconceptions about those
• 21:44
  autonomous agents? And from my experience, I used to be
• 21:48
  a management consultant a couple of markets and what I always have in mind
• 21:52
  is night trading. Like, like the Knights in
• 21:56
  Armor Night Trading group there was 2012,
• 21:59
  bankrupted by an algorithm going, going crazy. Within a
• 22:03
  day, they lost almost half a billion US dollars. And
• 22:08
  I'm just waiting for the headline news that an autonomous agency
• 22:12
  has also done this for a company making the bankrupt. But
• 22:16
  before we get into that, because you should not simply say,
• 22:20
  okay, let's work, it'll all be fine. You need to manage
• 22:23
  that. Do you see as the biggest
• 22:26
  misconception about autonomous agents
• 22:30
  in general? I think the biggest misconception is
• 22:33
  probably around the fact that we are already There
• 22:38
  to have true and 100%
• 22:41
  autonomy in its purest sense. When we
• 22:45
  speak about autonomy, I think there are certain degrees of
• 22:49
  autonomy. There's a spectrum of autonomy in
• 22:52
  executing certain tasks, right? I think on the very
• 22:55
  far end of the spectrum there's this human, human autonomy, right? If you have
• 22:59
  a worker and they get the task to get the best deal
• 23:03
  on supplying a certain material, then they would go ahead.
• 23:07
  They first create a plan, okay? I need to find out who are all of
• 23:10
  the relevant suppliers. Then I make a plan for the negotiation. I
• 23:13
  have them sent over test samples, right? I do the quality check.
• 23:17
  I in parallel need to check with accounting to set up the vendors in the
• 23:21
  ERP, etc. It's a multifaceted approach, right?
• 23:25
  With different systems interacting with different scenarios that
• 23:29
  you have to go through. This is the far end of autonomy, right? But if
• 23:32
  you start to cut it down and take away some of these
• 23:37
  dimensions and put it into one process, which is, for
• 23:40
  example, there are orders incoming, right? There's communication
• 23:44
  incoming. That communication has your products,
• 23:47
  it has a delivery date, it has an address, then this can also
• 23:51
  be automation, sorry, it can also be automation, autonomy, but
• 23:55
  in a far less complex way, right? And I think the
• 23:58
  misconception that is, that is very dominant still out there is
• 24:02
  we are at this far end that AI is possible
• 24:06
  to do what a human can do in all its facets and think about
• 24:10
  all of the possibilities. But what we see when you
• 24:13
  really apply AI and LLMs in a company
• 24:17
  in production scale, then what works right now is
• 24:21
  this, what I mentioned at the other end of the autonomy spectrum, right? You
• 24:25
  have to give the AI AI guardrails. You have to
• 24:29
  ensure that whatever should be deterministic, if
• 24:32
  then remains deterministic and is not overruled by an
• 24:36
  LLM because of its probabilistic nature, right?
• 24:40
  And I think that this misconception of we are already there,
• 24:44
  I think that's very dominant and it's a little bit dangerous because the
• 24:47
  expectations for LLMs and what they can do are completely
• 24:51
  overhyped, right? If you really want to have LLMs and
• 24:54
  AI work at scale, start with this beginning. Start with these
• 24:58
  processes that are well defined. Start with these processes
• 25:02
  where also humans exactly know what to do. And it's repetitive
• 25:06
  and it's clear we are not there yet at this
• 25:09
  very, very far end of, you know, full and true and 100%
• 25:13
  autonomy in all of its facets.
• 25:18
  I totally believe so. There are still a lot of
• 25:22
  jobs. You think, oh, the employees know what to do, but Then
• 25:26
  you realize at one point, well there's this aspect
• 25:30
  that it's not totally defined and we need to do decision. Well, there's
• 25:34
  this one, there's that one. You need to know X, you need to know Y.
• 25:37
  So even simple tasks are
• 25:42
  difficult sometimes, especially if you need to teach
• 25:46
  it through to machine.
• 25:50
  We go later into our founders world and they you'll talk about
• 25:54
  the one operational failure that kept you up at night and
• 25:58
  an AI employee would have helped to prevent.
• 26:02
  Let's talk a little bit about obstacles and how to overcome
• 26:05
  them. So we already talked a little bit about challenge,
• 26:09
  how far AI is. But what was the hardest technical
• 26:13
  operational challenge in building AI
• 26:16
  employees and how did you solve it? The biggest
• 26:20
  challenge really is to grasp
• 26:23
  what I would call context. Because
• 26:28
  if a process stays within the defined boundaries and
• 26:32
  everything is plain vanilla, then it
• 26:36
  is not as challenging to, you know, build the
• 26:39
  automation, run the automation. But as you mentioned, these exceptions are
• 26:43
  what makes it really, really interesting. And behind
• 26:47
  these exceptions still there can be rules. But these
• 26:50
  rules are typically just in the heads of those people that execute the
• 26:54
  process, right? So if I go back to the example of receiving
• 26:58
  orders, let me say this is a company that produces some food
• 27:01
  product and there are smaller companies that order from them, retailers
• 27:05
  or other small companies, then
• 27:10
  the context that may be in the head of the operators who previously
• 27:13
  received those orders and put them into, into the system, they
• 27:17
  may know, you know, a certain customer has certain
• 27:21
  conditions for the order, certain delivery days. You can only deliver on
• 27:25
  Mondays, Tuesdays and Wednesdays. These kinds of things typically are
• 27:29
  not encoded in the erp. They are not encoded in some file,
• 27:32
  they are just in the hands of these operators, right? Or for
• 27:36
  a certain customer, you have to interpret their language slightly differently,
• 27:40
  they refer to a product differently, right? And that context,
• 27:44
  that is really what drives complexity and also
• 27:48
  what makes it hard for LLMs and AI
• 27:52
  if you don't teach it properly to work at that scale.
• 27:55
  Because it becomes very frustrating for the team member, right. If it always have
• 27:59
  to correct the AI for the same exact thing, right? The
• 28:03
  AI over time needs to understand that context,
• 28:06
  these very, very special things that again only sit in the head of the
• 28:10
  operators. And that is something that, that we came across very early in our
• 28:14
  journey and that we solved with what we call our context
• 28:17
  engine. It's a part of the product that is absolutely critical
• 28:21
  to drive the efficiency and drive the
• 28:25
  accuracy of ADA over time because it
• 28:28
  captures that context that once again sits in the head
• 28:32
  of the operators exclusively in the head of the operators.
• 28:36
  And that was a real unlock for us to make sure that this
• 28:39
  context gets enriched, codified and used
• 28:43
  for the, for the operations going forward.
• 28:47
  Yes, that, that knowledge that is locked up
• 28:51
  in the mind of an employee, worst, worst case of
• 28:54
  a former employee, it's good when they retired because they like, like
• 28:58
  to come in again and talk about their job. But if they got
• 29:02
  fired, you won't get that knowledge back. So I
• 29:05
  totally understand what you're going for. You
• 29:09
  also bump if you do process optimization and some stuff so
• 29:13
  that there's always a lot of stuff that's not written down.
• 29:17
  Without naming confidential clients. What type of
• 29:21
  workflows are your employees already replacing
• 29:24
  today? The use cases
• 29:28
  really come from all parts of the organization. A lot of
• 29:32
  use cases, of course are in procurement, in the procure to pay
• 29:36
  cycle, but also in order to cash on the distribution
• 29:39
  side, in invoice
• 29:42
  processing, booking of invoices, you
• 29:46
  can think about applications even in other
• 29:50
  areas. So what we build
• 29:53
  is again customized. It fits to
• 29:57
  the specific process at the customer and it can be as
• 30:01
  easy as processing order confirmations
• 30:05
  from your suppliers, checking any differences and
• 30:09
  putting that into the ip. It can be part of your
• 30:14
  sales process. Right. If you think about a company that has
• 30:17
  large B2B customers that do an annual review of the
• 30:21
  contracts and discount negotiations, there's
• 30:25
  also a use case that we are, that we've built where
• 30:28
  the sales team interacts with ADA
• 30:32
  requesting the confirmation for certain discounts they
• 30:35
  want to offer. ADA does the calculation of whether this is within
• 30:39
  range and then gives the green light to the sales reps to go ahead and
• 30:43
  and interact with the customers and offer them these terms. Right. That is also
• 30:47
  a use case that we've built. There is a use
• 30:51
  case of matching incoming goods
• 30:54
  against the purchase order and against the invoice that was
• 30:58
  issued. Right. Which would be a three way match or you can even extend to
• 31:02
  a four way match. Right. Also, a previously human process
• 31:05
  of combining all of these data points, validating
• 31:09
  all of these documents and then acting in case of a
• 31:13
  discrepancy. That's also a use case that we've built. So there are many, many different
• 31:16
  use cases across the entire supply chain.
• 31:20
  And beyond that you can automate nowadays.
• 31:25
  What always comes to mind for me, keep in mind, I have a finance
• 31:29
  background, I work there a lot, is accounting processes.
• 31:33
  Because the funny thing is I once talked to
• 31:36
  a friend, if you're creative and you're creative in finance,
• 31:40
  you'll be rich. If you're creative and you are creative in
• 31:43
  accounting you'll go into jail. Right. So
• 31:47
  there are very certain limits that you, very
• 31:51
  strict rules that you have adhered to. But I do believe
• 31:55
  the potential that something goes wrong. There is also one of the
• 31:59
  reasons why we'll see this likely later on.
• 32:02
  Talking about tax filings. Yeah. If you have a
• 32:06
  mistake in there and usually you need to pay
• 32:10
  like a thousand euros a month in taxes and then it
• 32:13
  ends up 10 billion, you do have a problem.
• 32:19
  Yeah. Let's go a little bit into your playbook.
• 32:23
  What's like the biggest lesson for founders when
• 32:27
  identifying which workflows of theirs are
• 32:30
  ripe for agentic automation? And keep in mind, I think with the
• 32:34
  counter. Great guys. Very good
• 32:37
  point. No, when I think about
• 32:41
  founders, you
• 32:45
  should start thinking about LLM based or AI based
• 32:48
  automation when you have understood a process. I
• 32:52
  think that is a very good starting point because if you're still figuring out,
• 32:56
  if you're still testing, then you know, what are you going to
• 32:59
  automate? How do you going to tell the AI or the LLM
• 33:03
  or the system that you want to build what to do? Right. I think
• 33:07
  first comes the clarity of what's supposed to happen. How is it supposed to
• 33:11
  happen? How will it help me? And does it have scale? That really makes sense
• 33:14
  Because I tell you, building something that works
• 33:18
  reliably across, you know, all of the
• 33:23
  incoming signals that you can have, the variety of, you know, day
• 33:26
  to day operations takes time and takes effort. It
• 33:30
  may be reduced by now because there are all of these different tools that you
• 33:33
  can use, like you know, these Copilot Studios and N8N
• 33:37
  and all of these self serve environments which are certainly
• 33:41
  helpful, which you can certainly use. But you still need to have a
• 33:44
  good understanding of where you want to get to. Right. And that can only happen
• 33:48
  if you are clear about the process that you want to, that you want to
• 33:52
  automate. But if you have that, then I can encourage you to already start
• 33:56
  when you're building a company to already start very early on. Right. Because
• 34:00
  it certainly helps you to scale yourself and scale your time
• 34:05
  because you can move from doing these things to monitoring these
• 34:08
  flows and get more things done in a
• 34:12
  shorter time period.
• 34:16
  For our audience and the founders are listening, I was
• 34:20
  wondering think one repetitive process in your startup
• 34:23
  that breaks every week. Got it. Great. Keep that in mind as we
• 34:27
  move to the next section. Talking a little bit about
• 34:31
  your tactical framework here. Could you walk us through
• 34:35
  like 30 day deployment blueprint for onboarding
• 34:39
  a company's first AI employee. And
• 34:43
  when I think about onboarding and all the stuff you need to add in. Have
• 34:47
  you ever heard a politician talked about taxing AI
• 34:51
  employees? A very good
• 34:54
  question, taxing AI employees.
• 34:58
  No, I have not heard about that so far. But it's a very interesting
• 35:02
  thought, Very interesting thought.
• 35:06
  What's the playbook or what's the approach? So first,
• 35:10
  similar to my answer about, you know, when founders should think about
• 35:14
  AI based automation, you first need to have the use case clear,
• 35:18
  right? What is the process that currently creates a lot of pain, a lot
• 35:21
  of headache and ties up very valuable
• 35:26
  human resources, human capacities. And if you have that clear,
• 35:30
  then it's all about understanding that process in depth, really. Going
• 35:33
  through, going through the systems, going through the decision trees. Some of those
• 35:37
  may be obvious, some of those may be hidden. And you really only find
• 35:41
  out those rules when you go in, when you ask
• 35:45
  questions, and when you understand it in more depth. And
• 35:48
  we've really encountered quite a few times that having
• 35:52
  both a manager and the operator in the same meeting going through that
• 35:56
  process, some things emerge where
• 35:59
  both had discrepancies. Manager thought, it's handled this way.
• 36:03
  The operator says, no, no, no, we have to do it that way because
• 36:07
  of reason. Xyz, that happens quite often
• 36:10
  and that's a good thing because it creates clarity, it creates visibility.
• 36:14
  And very, very often we also have an element of
• 36:18
  changing business processes as we uncover them, as we go through
• 36:22
  them, the teams that are involved often agree to
• 36:25
  change certain things about the process because it makes the
• 36:29
  process better. We are not talking about automation, we're just talking about a
• 36:32
  shit process being a shit process. Sorry for my language. And a good process
• 36:36
  being a good process. And if you go into depth on these processes, you often
• 36:40
  find these shortcomings that you can fix right
• 36:44
  then and there, just changing how it's supposed to work and
• 36:47
  how the sequence should be. Now, if you have done that, if you have
• 36:51
  really understood what the process is, then we go ahead and we develop, right?
• 36:55
  We develop custom, we develop this bespoke for our customers. So we
• 36:59
  need a little bit of time, a few weeks to do that. But when we
• 37:02
  are done, we hand over the
• 37:05
  ready product. ADA is doing what ADA is supposed to do. And
• 37:09
  the customer can test from top to bottom. The customer can
• 37:13
  test on all different, you know, scenarios with
• 37:17
  live data. And we really appreciate feedback
• 37:20
  during that time because we can incorporate the feedback very, very quickly from one day
• 37:24
  to the next. And so over a short period of time, a week or two,
• 37:28
  we iterate up to a point where ADA creates a lot
• 37:32
  of value for the team and the team sees the value because the team has
• 37:35
  interacted with ADA a lot and sees that it can take over a lot
• 37:39
  of these processes and can help them really meaningfully in the day to day work.
• 37:43
  They really feel that they have less burden on their
• 37:47
  shoulders and then you're ready to go live. And that is
• 37:51
  typically a phase of 30 days. Maybe a little more, maybe
• 37:54
  a little less.
• 37:58
  I was, I was also putting myself in the shoes of those
• 38:02
  employees. And what do you know what's really cool? Because you can
• 38:06
  outsource the most boring Nerf and
• 38:10
  will breaking repetitive tasks to
• 38:13
  an AI agent that you have in your job
• 38:17
  and that usually helps. So I think there are a lot of people
• 38:21
  who are really interested in helping you there.
• 38:25
  Oliver, Normally we would go into an ad break, but we're
• 38:28
  recording for already more than 40 minutes. So
• 38:32
  I would suggest we make this part one and say
• 38:36
  goodbye and then have a part two. What do you say? Love
• 38:39
  it. Let's do it. Okay, guys. Oliver, thank
• 38:43
  you very much. We will be back with part two with
• 38:47
  Oliver and ADA AI.
• 38:54
  That's all, folks. Find more news streams,
• 38:57
  events and
• 38:58
  interviews@www.startuprad.IO.
• 39:03
  remember, sharing is caring.